Using a design-to-test capability for LTE MIMO (Part 1 of 2) System-level simulation helps engineers gain valuable insight into the design sensitivities of Long Term Evolution (LTE) Multiple-Input Multiple-Output (MIMO) systems By Greg Jue and Dingqing Lu, Agilent Technologies Long Term Evolution (LTE) Multiple-Input Multiple-Output (MIMO) technology has the potential to increase data rates for a single user. It does this by using multiple antenna techniques to transmit multiple and different streams of data. Consider, for example, that for 2x2 downlink Frequency-Division-Duplex (FDD) MIMO with a 64QAM (quadrature amplitude modulation) modulation depth, a potential peakdata rate of up to 172.8 Mbps can be achieved. Increasing the number of antennas for 4x4 downlink FDD MIMO with 64QAM increases the potential peak downlink data rate to 326.4 Mbps [Reference 1]. These data rates represent the upper boundary of what might be achieved with perfect radio conditions. In practice, however, MIMO performance is impacted by baseband and RF signal impairments from the system implementation, as well as MIMO channel conditions. MIMO system performance may vary as a function of baseband algorithm performance for receiver channel estimation, RF crosstalk coupling between antennas, RF/IF receiver phase noise, or RF/digital receiver analog/digital (ADC) clock jitter. Given the design complexities of LTE MIMO, it is a challenging task for the system engineer to partition RF and baseband system-design requirements (e.g., error and power budgets) to meet LTE specifications without over-designing. Re-working designs late in the testing phase can be costly in terms of both development cost and time-to-market. To help mitigate risk, it is critical that the engineer have visibility into potential issues as early as possible in the product development lifecycle. As the design cycle transitions from the design phase to early R&D hardware testing, it is useful to test R&D hardware Device-Under-Test (DUT) prototypes under various conditions. This allows the engineer to determine a prototype s sensitivity to different types of impairments (e.g. bit error rate (BER) versus bias or signal-to-noise ratio (SNR)). System-level simulation can be especially useful in helping the system engineer gain valuable insight into the impact of various RF and baseband impairments on MIMO system performance, thereby enabling design requirements to be evaluated. It can also be combined with test equipment for MIMO R&D hardware testing.
Baseband design Baseband coding and decoding algorithms are key to LTE system BER and Block-Errorrate (BLER) performance. The performance of the receiver-channel estimation algorithm can drive MIMO system performance. One of the primary challenges associated with designing to an emerging standard such as LTE is that the standards are rapidly evolving and, as a result, are subject to interpretation. FPGA engineers, for example, must typically study the LTE specification and interpret the pseudo-code algorithms as functional or behavioral design requirements for their FPGA HDL code. While their HDL code may be optimized for performance, it can be useful to compare the FPGA and HDL 0 s and 1 s, or vectors, against an independent reference to check for any misinterpretation of the LTE specification. Sometimes, FPGA engineers create their own test-vector references, in addition to writing their own HDL code. This is analogous to a writer creating their own spell checker any spelling misinterpretation may also be present in the spell checker. When writing algorithm code for an FPGA or DSP implementation by hand, LTE algorithm reference vectors, such as the one discussed in Reference 2, can be useful as an independent check of the LTE specification. Alternatively, the system engineer may want to evaluate the fixed-point precision needed for an LTE MIMO design for baseband-rf requirements partitioning. Consider the example in Figure 1. Here, the effective fixed word length of Finite-Impulse-Response (FIR) Root Raised Cosine (RRC) filters is varied to determine the fixed-point precision needed to achieve a MIMO Error Vector Magnitude (EVM) system design metric. EVM provides a convenient, single number metric to measure waveform quality and is typically specified for transmitter performance.
Figure 1. The results of a MIMO, fixed-point IQ modulator simulation that was accomplished using the Agilent VSA software. (Note that VSA software, version 11 or higher, can be used for the LTE MIMO analysis.) The simulation in Figure 1 consists of a MIMO downlink source which outputs complex MIMO data streams for 2x2 MIMO. The MIMO data is upsampled, RRC filtered and digitally modulated on a digital IF carrier using a Fs/4 carrier-multiplexing technique. The two streams of data are then fed into an Agilent 89601A VSA software simulation element to perform the simulation analysis. Simulation results show the effect of varying FIR RRC word length on system performance. The VSA display on the upper right shows the results with the effective FIR word length set to 10. The 64QAM constellation looks good and EVM is approximately 0.5 percent. The spectrum and EVM spectrum (EVM vs. subcarrier) also look good. The VSA display on the lower right, however, shows significant degradation to the system performance as a result of reducing the effective FIR word length to 7. The 64QAM constellation also shows distortion, which is reflected in the EVM increasing to 2.9 percent. The spectrum and EVM spectrum are impaired as a result of the fixed-point impairments in the IQ modulator design. With baseband impairments from the fixedpoint design impacting RF performance such as EVM and spectrum, design visibility can be especially useful in helping the engineer to perform system-level design trade-offs when partitioning baseband and RF design requirements.
For fixed-point baseband designs, it can also be useful to prototype the design with an FPGA development board. The engineer can then physically test the rapid prototype with test equipment. For the rapid prototype in this example, a modified version of the fixedpoint IQ modulator design was implemented (Figure 2). The MIMO IQ data was stored in Look Up Tables (LUTs) on the FPGA prior to upsampling, FIR RRC filtering and digital modulation of the IF carrier. HDL was then generated from the system simulation tool and FPGA synthesis was performed on it using a synthesis tool. Figure 2. This FDD MIMO FPGA implementation is a modified version of the fixed-point IQ modulator design. The FPGA development board contains two digital-to-analog converters (DACs) that convert the digital FPGA MIMO signals to analog. The two analog MIMO signals are fed into channels 1 and 3 of an oscilloscope and are demodulated by VSA software installed in the oscilloscope. Note that this is the same VSA software which was also used in simulation to design the fixed-point IQ modulator. In simulation, the VSA software processes simulated, instead of measured, signals from test equipment hardware. Using the software for both purposes ensures consistency in the measurement algorithms between design and test. The signal source below the oscilloscope provides a Continuous Wave (CW) clock signal to clock the FPGA board DUT.
RF mixed-signal transmitter design RF impairments such as antenna cross-talk, phase noise and power amplifier (PA) gain compression can also impact MIMO system performance. The local oscillator (LO) phase noise is of special interest in Orthogonal Frequency-Division Multiple Access (OFDMA) systems because low phase is important in maintaining subcarrier orthogonality, although it can be relatively expensive to achieve. Because the LO phase noise can be set in terms of dbc/hz at different frequency offsets, it is easy to simulate the actual phase noise of different LO solutions. The PA nonlinearity is evaluated by setting the 1-dB compression point on the two PAs at the output of the MIMO transmitter. To better understand the impact of RF impairments, consider the dual-channel RF transmitter shown in Figure 3. To the right of the figure is an antenna cross-talk model that was constructed to simulate the effective crosstalk coupling between channels 0 and 1. MIMO performance can vary as a function of the correlation between the multiple streams of data. Typically with RF 2x2 MIMO design, coupling occurs between the two antenna channels. Figure 3. This dual-channel RF, mixed-signal transmitter shown is designed using bandpass filters, mixers, LOs, and PAs to upconvert the IF from the fixed-point IQ modulators to RF. Figure 4 shows the simulation results with -80 dbc/hz phase noise at a 10-kHz offset and -20 db of antenna crosstalk. The output 1-dB gain compression point is set on the two power amplifiers.
Figure 4. Shown here are simulation results based on a -80 dbc/hz phase noise, PA gain compression and -20 db of antenna crosstalk. The VSA simulation result shows the composite EVM at 3.8 percent from the LO phase noise, PA gain compression and antenna crosstalk being modeled. Composite EVM includes Physical Downlink Shared Channel (PDSCH) data, primary and secondary sync channels and other control channels. The EVM spectrum (on the upper right of the VSA display) shows significant frequency variation from the frequency response of the coupled antenna path being modeled with the antenna crosstalk subnetwork. The Reference Signal (RS) EVM is approximately 1.7 percent EVM and is lower than the 3.8 percent composite EVM previously discussed. MIMO reference signals are orthogonal in both time and frequency, so RS EVM is typically not impacted by the cross-coupling of the two antennas. In contrast, the composite EVM is impacted by antenna cross coupling. By comparing composite and RS EVM, the system engineer can gauge how much of the EVM budget is being dictated by antenna crosstalk, versus other impairments present on the signal (e.g., phase noise and PA gain compression). Impairments can be further isolated to see what their contribution is on EVM. Figure 5, for example, shows the simulation results that occur when the LO phase noise and PA gain compression are removed, further isolating the effects of standalone antenna crosstalk on MIMO performance.
Figure 5. These simulation results occur with only -20 db antenna crosstalk, no LO phase noise and no PA gain compression. The antenna crosstalk is left at -20 db, but in the absence of phase noise and PA gain compression, the RS EVM drops to 0.05 percent. Due to the orthogonality of the two MIMO streams, the RS EVM is not sensitive to antenna crosstalk. The composite EVM drops to 3.3 percent since there is no phase noise or gain compression, only crosstalk. The close-up display of one of the constellation points on the right of Figure 5 helps the engineer gain insight into this antenna crosstalk impairment. There is a Constellation of Constellations resulting from a small 64QAM crosstalk signal being summed in with the main through path. Previously, the phase noise and gain compression present on the signal masked this cross-coupling effect. System engineers can quickly and easily gain insight into various design sensitivities by modeling different impairments in simulation and trying what-if scenarios. This provides the engineer with a valuable tool to gain insight into design performance and ensures that informed design trade-offs can be made when specifying design requirements. As a result, the engineer is better able to mitigate risk and avoid overdesigning. Part 2 will look at receiver RF/mixed-signal design, and R&D hardware DUT testing.
References 1. Agilent Technologies (publisher), LTE and the Evolution to 4G Wireless: Design and Measurement Challenges, (2009). Edited by Moray Rumney,Table 1.4-1, ISBN 978-988- 17935-1-5 2. Agilent Technologies, White Paper on LTE PHY Design, http://www.agilent.com/find/eesof-lte-whitepaper. About the Authors \Greg Jue is an applications development engineer/scientist with Agilent EEsof Electronic Design Automation (EDA), specializing in SDR, LTE and WiMAX applications. Greg wrote the design simulation section in Agilent s new LTE book, and has authored numerous articles, presentations and application notes, including Agilent s new LTE algorithm reference whitepaper and Agilent s new Cognitive Radio whitepaper. Greg pioneered combining design and test solutions at Agilent Technologies, and authored the popular application notes 1394 and 1471 on combining simulation and test. Before joining Agilent in 1995, he worked on system design for the Deep Space Network at the Jet Propulsion Laboratory, Caltech University. Dingqing Lu has been with Agilent Technologies/Hewlett Packard Company since 1989 and is a scientist in Westlake Village, CA, USA. From 1981 to 1986 He was with University of Sichuan as Lecturer and Assistant Professor. He was a Research Associate in EE Department of UCLA, Los Angeles, USA from 1986 to 1989. He published about 20 papers in IEEE Trans, Journals and Conference proceedings. His research interests include modeling, simulation and measurement techniques for systems.